Print 15 comment(s) - last by Argon18.. on Dec 19 at 11:59 AM

Intel better watch its back

According to Bloomberg, Google is considering designing its own server chips in an effort to form a more fluid interaction between its hardware and software. Google will reportedly use technology from ARM Holdings Plc. 
Google is currently Intel's fifth largest customer, but if the search giant were to move into server chips, that could hit Intel's revenue. Bloomberg said Google represents 4.3 percent of Intel's revenue. 

It would make sense if Google started designing its own server chips, considering it has already designed its own data centers complete with servers around the globe. Working with ARM isn't a bad idea either, since its chips currently rule the mobile sector when it comes to smartphones and tablets. 
There are a couple of clues that point to Google's interest in server chips. For starters, the tech player recently joined a group started by IBM, which licenses technology used in data centers -- such as chips for servers.
Google also recently posted a job opening for a digital design engineer. Specifically, Google wants someone with qualifications in application-specific integrated circuits (ASICs).
Google has not yet confirmed the chip-designing rumors. 

Source: Bloomberg

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

Good ol' Google.
By KurgSmash on 12/13/2013 3:22:56 PM , Rating: 2
Their motto should be "do 500 things, do 10 of them right."

This would be an epic fail if they do try it. ARM in the data center is foolish.

RE: Good ol' Google.
By Spuke on 12/13/2013 7:28:08 PM , Rating: 2
ARM in the data center is foolish.
So foolish it's going to happen.

RE: Good ol' Google.
By Shig on 12/14/2013 1:12:59 PM , Rating: 2
I stopped questioning Google after their stock price passed 1,000$.

RE: Good ol' Google.
By Solandri on 12/14/2013 2:11:34 PM , Rating: 4
Their motto should be "do 500 things, do 10 of them right."

That is precisely why Google succeeds.

Most people think that they can predict based on theory what will or will not succeed. Google figured out long ago that even very smart people are terrible at making those sorts of predictions. So they front the money to actually try the idea for real first, then decide whether or not it's worth pursuing.

Google isn't the first to do this either. Thomas Edison tested thousands of materials for his light bulb before settling on a carbon filament. No biases, no preconceptions; try it out and see how it does. In fact Edison's greatest failure came about when he refused to even try an AC power distribution system because he was convinced that hypothetically the danger of electrocution outweighed any benefits.

You learn as much from your failures as you do from your successes. The tinkerer, the innovator, the pioneer will proudly wear their failures on their sleeve. The salesman OTOH only wants his successes publicized.

RE: Good ol' Google.
By morgan12x on 12/14/2013 8:31:16 PM , Rating: 2
Notice that they are looking for someone with experience in ASICs. Not ARM specifically. ASICs are built to do one thing, do it really fast, and really efficiently. (Search for Bitcoin ASIC mining) All they have to do is build chips to do SQL searches, or content delivery, etc. A one-off design to do one thing really well. Heck, they can even build them using FPGAs and change up the design later.

RE: Good ol' Google.
By drycrust3 on 12/15/2013 2:22:44 PM , Rating: 2
Google figured out long ago that even very smart people are terrible at making those sorts of predictions. So they front the money to actually try the idea for real first, then decide whether or not it's worth pursuing.

Innovation is all about at least considering new ideas, you don't just ignore a problem, you try and find a solution, or if you can't fix it then try and find a partial fix.
In the case of "do 500 things, do 10 of them right", I think that doesn't tell the true story. My experience is that for every 100 suggestions there are 5 to 10 worth another look at, so if Google tried 500 things then employees made at least 2500 suggestions on how to fix problems or new ways of using new or existing technology, which is an excellent amount ideas. I think the "10 done right" is probably a bit under representative of the effect of 500 trials, although it could represent the very best, so those 10 don't just represent 10 innovative ideas, they represent 10 new types of technology that benefit mankind.

RE: Good ol' Google.
By woody1 on 12/15/2013 10:16:28 AM , Rating: 2
Google must be kicking themselves for not consulting you before launching this initiative. They probably deluded themselves into thinking they had some really smart engineers and just didn't realize that you were out there, and how much more clever you are than there people.

RE: Good ol' Google.
By Arkive on 12/16/2013 4:28:05 PM , Rating: 2

Google has prescribed to the notion of try a lot of things, hone the ones that show merit, and of those market the ones that shine (contrary to Apple who has taken the opposite approach of make a few things perfect and ignore almost everything else). I'm not saying one is better than the other, but to say Google will fail simply because ARM isn't a recognized component in the data center is ridiculous. If it wouldn't work they wouldn't be humoring it, unless this is just some ploy to freak out Intel and get them to offer them a better price on processors (which seems too obvious to be true).

RE: Good ol' Google.
By Argon18 on 12/19/2013 11:59:35 AM , Rating: 2
"This would be an epic fail if they do try it. ARM in the data center is foolish."
Only a Redmond-worshiping Wintard would make this claim. Since of course Microsoft has no server product for ARM, and their consumer products for ARM are all market failures.

x86 v ARM costs
By Shig on 12/13/2013 2:58:48 PM , Rating: 2
Developing your own ARM chip is ~30-50 million dollars. Developing your own x86 chip is about 300-400 million dollars.

RE: x86 v ARM costs
By bsd228 on 12/13/2013 3:10:56 PM , Rating: 3
The cost I expect Google is driven by is the operating costs of their datacenters, with millions of servers and thus millions of cpus.

but Intel has been doing very well at lowering their cpu power draws to the point where for most the ARMs don't make sense. They do much less work and you still have the power overhead of the rest of the compute unit, and you have the manageability issues of having more, less powerful, nodes. Google is fine on the latter problem - they already manage a ton of nodes.

RE: x86 v ARM costs
By 91TTZ on 12/13/2013 5:01:40 PM , Rating: 2
That's not a fair comparison. It costs less to develop an ARM chip because ARM is a fabless chip designer that licenses its designs to you, and you're merely customizing the design. They're actually doing the hard part and they get paid for it.

On the other hand Intel doesn't license its IP to anybody, and you'd probably get sued if you tried to come out with your own x86 chip.

Will it be scaleable?
By Shadowself on 12/14/2013 7:21:31 PM , Rating: 2
If Google does a "server ARM chip" of their own design based upon ARM's instruction set, it will only be beneficial to them if they can scale it to their mobile devices. The cost/benefit of pouring money into designing your own chip just to save power costs of the CPUs themselves does not make sense. CPU power usage in a huge data center is a small fraction of the overall power requirements.

However, if they develop a true 64 bit chip for their servers AND can cost effectively create a true 64 bit chip for their mobile devices scaled from those server chip designs, then Google might really have something.

RE: Will it be scaleable?
By jamescox on 12/16/2013 8:16:58 AM , Rating: 2

4.3 percent of Intel's revenue is going to be a sizable amount of chips, but I don't see why they wouldn't use it for their mobile devices also.

Any chip they develop for mobile at this point in time will certainly be 64-bit capable. I am not sure what you mean by "true 64 bit" chip. It would have seemed crazy to put a 64-bit chip in a mobile device a short time ago, but we already have devices like the Galaxy Note 3 with 3 GB of ram, so 4 GB devices should probably be coming out in the next year or two.

The requirements for mobile and data center (most compute per watt) are actually not that far apart anymore. We are seeing more and more mobile chips which are limited by their power envelope, so scaling up to data center may just be turning up the clock speed. Maybe increase the caches and memory bandwidth also. Wider memory interfaces take more power though, and may not be worth it for server apps; these are usually more latency sensitive than bandwidth sensitive.

There doesn't seem to be that much evidence supporting that they are actually doing this though.

By marloloh on 12/15/2013 5:00:18 PM , Rating: 2
if you need a job try this site JOBS61 (dot)¢øm. Dan does it at home and makes $29.98 hourly just sitting and typing stuff all day...No experience needed too

"I'd be pissed too, but you didn't have to go all Minority Report on his ass!" -- Jon Stewart on police raiding Gizmodo editor Jason Chen's home

Most Popular ArticlesTop 5 Smart Watches
July 21, 2016, 11:48 PM
Free Windows 10 offer ends July 29th, 2016: 10 Reasons to Upgrade Immediately
July 22, 2016, 9:19 PM

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki